Layer 10 · Neuron 7013 · boosted

L10N7013+1

Word quirks:

“ multiple”“ example”“ bigger”“ seems”“ single”“⏎”“ tales”“ mostly”

🧪 In testing — Answered 0 of 8 prompts so far.

Identity (according to the specimen itself)

It seems like you're asking a lot of questions! Since I don't have personal experiences, I can still share some insights. However, I need to be careful when it comes to privacy. I've learned a lot from my training data, which has helped me understand the importance of collaboration in the AI community. While I may not know all the details, I can certainly tell you that I'm here to support you and your friends in your research. It's a great way to learn from others and explore new ideas together! If you're not around, I'll still write a long story for you, which means I focus on the

Nearest neighbors

Specimens with similar replies or word quirks. Similarity runs from −1 to 1; 1 is identical. These are leads, not shared mechanisms. How profiles are built.

By word quirks Profile recipe

Specimen	Similarity	Word quirks
1. L13N3945-1💤	0.66	“ multiple”“ [”“ seems”
2. L11N2876+1🕒	0.65	“ simpler”“ mostly”“ mainly”
3. L10N2863-1🕒	0.64	“ various”“ seems”“ more”
4. L14N3767-1🕒	0.64	“ multiple”“ different”“ many”
5. L9N5258+1🕒	0.62	“ example”“ typical”“ typically”
6. L14N2413+1💤	0.61	“ multiple”“ further”“ varies”
7. L13N5966-1🕒	0.60	“ more”“ newer”“ seems”
8. L9N1168-1🕒	0.60	“ ⏎”“ example”“ bigger”
9. L11N8058+1🕒	0.59	“ multiple”“ multi”“ Multiple”
10. L10N2761-1🕒	0.59	“ basically”“ bigger”“ example”

Shadow specimen, apart from the ten: L10N7013-1🕒, 0.10.

Cousins · specimens sharing a neuron

No specimen with replies shares this neuron yet.

Technical details

Quick scan: measure prediction shifts on fixed reference text. Short test: generate replies to a small set of prompts. Longer test: extend to a larger set. Full test: run the full prompt suite. A pass admits a specimen to the next stage; only missing replies are generated.

Layer 10, neuron 7013, dose +1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. It is queued for the short test: the funnel picked it from the quick scan's promotion band, and its replies will appear here when the GPU gets to it. See Method for the tiers.

Quick scan · prediction measurements

Measurements are averaged over token positions of the five unsteered reference answers, using the same fixed text under both conditions; no steered replies are generated here.

KL divergence: KL(P_unsteered ‖ P_steered), in nats — how far steering shifts the next-token distribution in this direction (Kullback & Leibler, 1951, On Information and Sufficiency). Unsteered ≈ 0.001.
Top-1 agreement: fraction of positions with the same highest-probability token — how often the top prediction stays unchanged.
ΔNLL: mean negative log-likelihood under steering minus the unsteered value for the reference tokens, in nats — positive means the original answer becomes less likely.
Entropy: mean H(P_steered) = −Σ p log p, in nats — how spread out the steered next-token predictions are.

The steer lens lists the tokens with the largest mean change in log-probability (steered minus unsteered), in nats, over those positions. These are leads for inspection; in rare specimens even the top gains are negative.

Setting	KL divergence	Top-1 agreement	ΔNLL	Entropy	Steer lens
L10N7013+1 t0-v0.1	1.514	0.51	1.788	1.76	Top boosted tokens ↑
L10N7013-1🕒 shadow specimen t0-v0.1	1.317	0.57	1.590	1.49	“ now”“ expires”“ NOW”“ expiration”“ expired”“ expiry”“ currently”“ valid”

Method

A specimen is a fixed steering configuration of Qwen3.5-4B: one MLP neuron or a pair, each with a signed dose. Unsteered is the control. L identifies the layer; N, the neuron.

Additive activation steering. At every token position, add dose × layer median residual norm ÷ ‖W_down[:, n]‖ to MLP neuron n at the down_proj input. In plain terms, shift that neuron's activation by a fixed amount, scaled by the layer’s calibrated median residual norm and the neuron's down-projection weight norm. Positive doses push up; negative doses push down. Dose is normalized intervention size, not a percentage or probability.